Business Rules Forum 2003 Conference
November 2-6, 2003 – Nashville, TN
Rules-Based
Interaction Processes
Jacob Feldman, Ph.D.
Edison, NJ
1
Presentation Outline
• Dynamic Interaction Processes: 4 Scenarios
• Rules-based Interaction Library
– Dialog Construction Rules
– Dialog Execution Rules
– Integration
• Live Demo: Dynamic Loan Approval
2
Dynamic Interaction Processes
• In complex decision making environments
such as loan origination, tax compliance,
portfolio balancing, insurance underwriting, or
airport security the interaction logic is
dynamic by its nature and cannot be
predefined in advance
• Below are several examples (scenarios) from
real-world business environments that
illustrate the problem
3
Scenario 1: Loan Approval
#
Events and Facts
Peter Johnson’s
Loan Application
Decision
Comments
1.
Peter Johnson applied for $50K loan
Rejected
Insufficient Income. But a
bank manager found that their
valuable client with the same
address can be a guarantor
2.
Joe & Dawn Johnson agreed to be Peter’s
guarantors. They have a Housing Loan
with Available Equity = $300K and
Remaining Debt = $150K
Accepted
$100K surplus is a sound
proposition. But Conducting
more detailed analysis, the
manager notices a joint
borrowing on Mr Johnson file
which is not with his wife
3.
Joe Johnson and his partner Bill Smith
(50/50) have a Business Loan for $200K
with Available Equity $52K
Accepted
The remaining $26K surplus
still meets requirements. But
Bill and Susan Smith have
other facilities outstanding
against their property as well
as the business loan
4.
Bill & Susan Smith have a Housing Loan
with Available Equity = $240K and
Remaining Debt = $150K
Accepted
Still a surplus. But a lending clerk
at the lending operations center
while preparing the collateral
Their son Tommy Smith has $50K loan
secured by his parents
Rejected
5.
documentation, noticed a secured
personal loan in the name of Tommy
Smith for $50K secured by his
parents
Available Equity
$42K < $50K.The business
debt would be $8,000 short on
cover
4
Scenario 1: Raised Questions
• To solve a business problem like this one, humans
and machines should make an approve/decline
decision based on the joint performance of tasks
• While new facts about the loan related securities can
come from different sources, an effective interactive
system should be able to request new information:
– Based on the information entered for a particular case and
– Based on related existing knowledge about the customer.
• What does “related knowledge” mean? Where is it
defined and how can it be brought to the picture?
5
Scenario 2: Tax Return Compliance
for Partnerships
• Tax compliance becomes extremely complicated
when it has to consider a possible revenue and
distribution of expenses among multiple related
corporations (partnerships including foreign ones)
• The main problem here is how to generate the
proper data requests, receive the related data, and
then validate the accumulated information to be
compliant with the current regulations
• Again, we deal with a question of how to define and
maintain the “related data”, which is dynamic in
nature and can not be saved in a database
6
Scenario 3: Maintenance of User
Profiles for Portfolio Balancing
• A customer may define preferences related to his/her
investment strategy (conservative or moderate risk
level, industry sectors, security type distributions, etc.).
However, the dynamic nature of the constantly
changing financial market requires permanent
automatic and interactive adjustments to each
customer’s profile
• For example, a system should be able to generate
questions like: ”Your positions are overly concentrated
in a single security. Are you willing to relax position
constraints?” and make an automatic decision in each
case based on a customer’s preferences and the
company’s current strategy
7
Scenario 4: Identifying Suspicious
Groups of Airplane Passengers
• A system validates a list of all passengers when they book
tickets for air travel. Along with simple criteria such as:
• age range, gender, country of origin, legal status, etc.
the system may include dynamic characteristics such as:
• acquired certain chemical products in certain quantities,
• took certain classes at a certain educational institutions in a certain
time period,
• visited certain countries during the last 2 years, etc.
• Dynamic attributes need to be validated not just for one
passenger but also for all possible combinations of
currently known passengers. The very fact that a passenger
satisfies a certain criterion, may initiate a new request
about other passengers, that can in turn initiate additional
new requests or reconsider already known facts
8
The Common Thread
•
•
What do all these scenarios have in common?
They all deal with the joint performance of
tasks by humans and machines:
– with the dynamic structure of these communications
– when the interaction logic can not be predefined and
depends on the interaction history and the related
information about involved objects
– with the necessity to support multiple data information
requests, algorithms and programming interfaces
•
We have to deal with situations when not all
concepts/relationships are known in advance
and new concepts/ relationships could be
added as we go
9
A Systematic Approach
•
•
The identified problem is very broad. Its
solution requires a systematic long-term
research and development activity that will
probably combine several technologies
including:
– Ontology/RDF/Semantic Web
– Rules-based techniques (BRE)
– Constraint satisfaction techniques
Many standardization bodies already started to
consolidate domain-specific knowledge repositories
(e.g. Basel II, ACORD, MISMO). Thus, methods and
tools to work with the consolidated information will be
required
10
A BRE Approach
• We tried to limit the scope of the
problem to provide practical solutions in
today real-world environment, and, of
course, BRE was our first choice
• In the remaining part of the presentation,
we will:
– describe a practical approach to dynamic
interaction using BRE technology
– demonstrate the approach using the Loan
Approval scenario
11
BRE & Dynamic Interacting
•
•
While business rules technology has already proven its
effectiveness for many practical applications with
complex business logic, it cannot be directly applied as
a solution to problems that require dynamic interacting
We have to address the following problems:
1) Can a rules-based system construct an interaction process
“on the fly” based on the interaction history and
information already available in a particular business
context?
2) How to apply business rules to define multiple
customizable dialogs with a variable content and complex
inter-question relationships?
3) How to achieve it and not to be lost in the complexity of
the supporting business rules?
12
Required Functionality
• Ability to ask the next question based on the
previous answers and a problem specific data
• Questions should be generated in a form that is
understandable either by a human or by a
computer
• Support for typical interaction constructions like
multi-choice questions, auto-responses, interquestion relationships
• Integration with different business contexts
(information sources)
• Integration with different portal environments
13
High-Level Architecture
External
Data Sources
Domain-Specific Knowledge Repository
(Ontologies,XML Schemas, DBs, Web Services)
Human
Analyst
Interaction Processor
Requests
Application
Input
Other ESR
Third
Party
Applications
Applications
Responses
Application
Rules-based
Interaction
Library
INTERACTION LOGIC:
1.
2.
3.
4.
Create All Possible Question Types
Generate Next Question based on the
Interaction History and Existing Data
Execute application-specific decision rules
Stop if no questions to ask
14
Rules-based Interaction Library
•
•
An extensible Interaction Library was created to
address the above problems using just business
rules technology and a GUI-based interaction
processor
The Interaction Library consists of two parts:
1. A Java package to support generic interaction
concepts and constructions
2. Set of templates that predefine semantics of the rules
for different dialog construction like multi-choice
questions, questions with answers from a certain
domain, events, messages, and auto-responses
15
Interaction BOM
• The Interaction Library is based on a generic
business object model (BOM) to support different
rules-based interaction processes
• The BOM is implemented in Java and covers such
concepts as:
– Interaction Session for a multi-user environment
– Interaction History to keep track of all asked questions
and provided answers
– Interaction Dispatcher to execute the interaction logic
– Interaction Processor to “ask” questions and receive
answers from GUIs, DBs, or other sources
– Questions and Answers
16
Interaction Java Package
•
•
Java API to an Interaction Rule Project
An application just creates an Interaction Session
that uses two predefined rule engines:
–
Rule Engine “Create Dialog”
–
–
Rule Engine “Execute Dialog”
–
•
creates interaction dialogs with all possible pages, sections, and
questions defined by application specific rules
called constantly during the interaction cycle to resolve interquestion dependencies with respect to the interaction history and
to define the next question to ask.
There could be also PreProcessing and PostProcessing
rule engines that contain application specific business
rules
17
Dialog Construction Rules
• The Interaction Library includes generic rules
common for any interaction processes
• The predefined rules are used to:
• Construct Dialog Pages
• Construct Sections on the Pages
• Construct Questions inside Sections:
– Multi-Choice Questions
– Questions with Answers of different controllable types
• Modify Questions based on the Interaction History
• Provide Auto-Responses:
– Respond to {question} using {string}
– Respond to {question} using {formula}, etc.
18
Question Construction Rules
(OpenRules implementation)
19
How to Define Question Types
• Using a very simple
XML file like this one:
or
an Excel spreadsheet
like this one:
• Or using another application specific way
20
Rules for Inter-Question
Relationships
• These business rules define the processing
logic based on answers to already asked
questions in conjunction with information
available from application specific contexts:
if answer to question <Q1> is <A1> and loan was
<declined> then ask question <Q2>
• We use decision tables for the interaction logic
to support complex inter-question relationships
• Decision tables allow us to combine both:
• Predefined interaction rules like:
“If answer to question <Q1> is <A1>” and
• Application specific rules like:
“If loan was <declined>”
21
Navigation Decision Table
(OpenRules example)
22
Rules-based Interaction Library
Live Demonstration
1) Web-based Loan Approval Interaction
- how does it look ( a user view)
2) Underlying Rule Project
- how was it done (an admin
view)
23
Demo Scenario: Loan Approval
#
Events and Facts
Peter Johnson’s
Loan Application
Decision
Comments
1.
Peter Johnson applied for $50K loan
Rejected
Insufficient Income. But a
bank manager found that their
valuable client with the same
address can be a guarantor
2.
Joe & Dawn Johnson agreed to be Peter’s
guarantors. They have a Housing Loan
with Available Equity = $300K and
Remaining Debt = $150K
Accepted
$100K surplus is a sound
proposition. But Conducting
more detailed analysis, the
manager notices a joint
borrowing on Mr Johnson file
which is not with his wife
3.
Joe Johnson and his partner Bill Smith
(50/50) have a Business Loan for $200K
with Available Equity $52K
Accepted
The remaining $26K surplus
still meets requirements. But
Bill and Susan Smith have
other facilities outstanding
against their property as well
as the business loan
4.
Bill & Susan Smith have a Housing Loan
with Available Equity = $240K and
Remaining Debt = $150K
Accepted
Still a surplus. But a lending
clerk also notices a secured
personal loan in the name of
Tommy Smith for $50K
secured by his parents
5.
Their son Tommy Smith has $50K loan
secured by his parents
Rejected
Available Equity
$42K < $50K.The business
debt would be $8,000 short on
cover
24
A Web-based Front End:
Loan Application
25
Rejected. Requests a Guarantor
26
Guarantying Security
(Joe&Dawn Johnson)
27
Related Security
(Joe Johnson &Bill Smith)
28
Related Security
(Bill&Susan Smith)
29
Related Security (Tommy Smith)
30
No Related Securities Anymore
31
Peter Johnson’ Loan Application
Declined ($8K less)
32
Demo Implementation
• Each time when a user answers to the
generated questions and moves to the next
page, the decision-making rule engine is
called
• Each engine’s run recalculates the total
equity and total remaining debt using ALL
related securities known so far, and then it
makes an Accept/Decline decision
33
Rule Administration Tool:
Loan Approval Rule Project
34
Integration with Business
Contexts
• Our approach moves the interaction logic into an application
specific rule project, so it becomes a BRE problem
• To get access to an external business context, the underlying
rule language should be able to deal directly with objects like
Driver, LoanApplication, or TaxReturn described
“somewhere” outside rule project
• The current implementation of our Interaction Library works
directly with:
– Java-based business objects
– XML Structures (without creating intermediate Java objects)
• We believe that a good rule language should be able to deal
with objects (Driver, Loan, etc.) in the same way whether
they come from Java, Database, XML schema, WSDL or
RDF
35
Integration with Portals
• While the presentation logic is usually outside of
business rules reach, the Interaction Library not only
defines the dialog pages, sections, and questions, but
can also set their presentation attributes like:
Layout=horizontal
Type=textarea
• A customizable build-in graphical interface allows a
developer to easily plug-in the interaction rule
project into different Web-based portals
36
Portal Integration Schema
Pages,
Sections,
Questions
Inter-Question
Relationships
Rules
Administration
Tool
Business Rules
Repository
Predefined Templates
Create Dialog
Rule Engine
Execute Dialog
Rule Engine
Rendering
XML
Page
Data
Web Client
(IE)
Web Client
(IE)
…
Web Client
(IE)
Interaction Library
Java Objects
Generated
HTML
XSLT
Processor
Web App Server
CSS file
XSL
Page Template
37
Important Observations
• An important point: rules generate requests
• A concrete Interaction Processor’s
implementation converts these requests to
queries or to GUI questions (as we did in this
demo)
• Ultimately, there could be no human
interaction at all, while knowledge “what and
when to ask” still can be kept in the same
rules
• A Web dialog of this demo is only one of
many possible interaction forms
38
Used BREs
•
•
The rules-based Interaction Library is BRE
vendor-independent. However, only rule engines
with a powerful decision table rules templatization
mechanism can be used
There are two current implementations:
1. Exigen Rules (www.exigengroup.com)
2. A new free Open Source full-scale BRE product
“OpenRules” (http://openrules.com)
Note. Other popular BREs can be added upon a request
39
Summary
• The proposed rules-based Interaction
Library provides a practical solution to
real-world business problems with
complex interaction logic
• Along with an ability to build powerful
interaction servers, it provides a
customizable Web-based front end for
dynamic dialogs
40
Questions & Answers
Jacob Feldman, Ph.D.
Edison, NJ
732-662-7233 office
732-306-0685 cell
jacobfeldman@openrules.com
41